Cell Scoring Neural Networks

This repository contains the source code used for CSNN.

Steps to run:

1. Download the data

2. Extract CLL_288.zip and B-ALL_178.zip to data/raw/: You should have the following structure in your data/ folder:

📦csnn
 ┗ 📂data
   ┗📂raw
    ┣📂B-ALL
    ┣📂CLL_24
    ┣📂CLL_102
    ┗📂CLL_162

3. Install the requirements: pip install -r requirements.txt

4. Run preprocess_ball.py

5. Run preprocess_cll.py

6. Run the experiments:

Experiment	Dataset	Algorithm	Command
Hyperparameter search	B-ALL	CellCNN	python train_cnn_cv.py config/tuning_ball_cellcnn.yaml
		DeepCellCNN	python train_cnn_cv.py config/tuning_ball_cnn.yaml
		CSNN-Class	python train_logistic_cv.py config/tuning_ball_logistic.yaml
		CSNN-Reg	python train_reg_cv.py config/tuning_ball_reg.yaml
	CLL	CellCNN	python train_cnn.py config/tuning_cll_cellcnn.yaml
		DeepCellCNN	python train_cnn.py config/tuning_cll_cnn.yaml
		CSNN-Class	python train_logistic.py config/tuning_cll_logistic.yaml
		CSNN-Reg	python train_reg.py config/tuning_cll_reg.yaml
Test set evaluation	B-ALL	CellCNN	python train_cnn.py config/best_ball_cellcnn.yaml
		DeepCellCNN	python train_cnn.py config/best_ball_cnn.yaml
		CSNN-Class	python train_logistic.py config/best_ball_logistic.yaml
		CSNN-Reg	python train_reg.py config/best_ball_reg.yaml
	CLL	CellCNN	python train_cnn.py config/best_cll_cellcnn.yaml
		DeepCellCNN	python train_cnn.py config/best_cll_cnn.yaml
		CSNN-Class	python train_logistic.py config/best_cll_logistic.yaml
		CSNN-Reg	python train_reg.py config/best_cll_reg.yaml
No initialization ablation	B-ALL	CSNN-Class	python train_logistic_ablation_no_init.py config/best_ball_logistic.yaml
		CSNN-Reg	python train_reg_ablation_no_init.py config/best_ball_reg.yaml
	CLL	CSNN-Class	python train_logistic_ablation_no_init.py config/best_cll_logistic.yaml
		CSNN-Reg	python train_reg_ablation_no_init.py config/best_cll_reg.yaml
Initialization only ablation	B-ALL	N/A	python train_ablation_init_only.py config/best_ball_reg.yaml
	CLL	N/A	python train_ablation_init_only config/best_cll_reg.yaml

System requirements

	Minimum	Used in this study	Notes
Processor	1 x86 compatible threads	8 Intel Xeon Gold 5218 threads	More threads makes the initialization run faster
Memory	16GB RAM	16GB RAM	Amount of RAM dependends on dataset size
GPU	1 CUDA compatible GPU	1+ RTX 2080ti or faster	GPU should have at least 11GB VRAM

Package versions used

Package	Version
torch	1.9.0
numpy	1.20.1
pandas	1.1.1
tqdm	4.61.2
matplotlib	3.3.1
pyyaml	5.3.1
scikit-learn	0.24.2